Built for the way you work.
The privacy filter replaces names, emails, account numbers, secrets, and other identifiers with semantic placeholders before a prompt reaches a closed model.
Zero Data Retention keeps prompts, code context, and completions in memory only. The response is returned, then the working data is discarded.
The same policy follows every product surface: Router, Enterprise Inference, CLI, Agents API, and VS Code extension.
What you get.
Identifier minimization before the model
PII anonymization removes the who from the prompt while preserving the task. Closed models receive placeholders such as <EMAIL_1> or <PERSON_2>, then the response is restored for your user.
The filter runs as a local inference step in the Blackbox layer, so identifiers are removed before provider routing. Enterprise teams can tune the categories that must be masked for their workflows.
Zero-retention request lifecycle
Prompts, code context, and completions stay in memory for the active request and are discarded after the response. No model training, no prompt review queue, no product analytics copy of the content.
Training opt-out is on by default, and we enforce it with the upstream providers that we route to, wherever the provider API supports it. For teams that need evidence, audit logs record access and configuration events without storing prompt bodies.
Sensitivity-aware routing
Use the Router for tasks that can reach closed providers with identifiers removed. Route regulated or crown-jewel code to Enterprise Inference, where Blackbox runs the open-weight model on capacity reserved for you.
Teams can separate public docs, internal tickets, source code, secrets, and regulated records by workspace policy instead of relying on every developer to pick the safest model manually.
Private inference starts before provider routing.
The privacy path reduces what each model sees, keeps transient data out of storage, and moves sensitive workloads away from shared provider infrastructure.
Classify the payload
The request is inspected for identifiers, secrets, account numbers, and private references that do not need to reach a model.
Replace identifiers
Matched values become placeholders that preserve meaning for the task without exposing the person, account, or secret.
Route by sensitivity
Closed providers receive minimized prompts; regulated work can route to a single-tenant open-weight deployment instead.
Discard working data
After the response is returned, prompt context and completion content are discarded under Zero Data Retention.
Privacy controls for teams that still need frontier models.
The route gives buyers a concrete answer to the privacy question: what changes before data leaves Blackbox, what is retained, and when closed models are excluded.
PII stripped before providers
Names, emails, phone numbers, URLs, account numbers, and secrets can be replaced with semantic placeholders before closed-model routing.
Zero Data Retention
Prompt bodies and completions are processed in memory and discarded after the response, so the working content is not reused as a storage layer.
Sensitive-workload isolation
Workspace policy can reserve single-tenant Enterprise Inference for source code, regulated records, or anything your team will not send to a third party.
Common questions.
IS PRIVACY DIFFERENT FROM SECURITY?
Yes. Security controls who can access the system and how data is protected. Privacy controls how much sensitive data reaches each model or provider. Blackbox uses both: encryption and access controls for security, plus PII minimization and Zero Data Retention for privacy.
WHAT DOES THE PII FILTER REMOVE?
The filter detects common personal and sensitive data categories such as names, email addresses, phone numbers, private URLs, dates, account numbers, and secrets. Matched values are replaced with semantic placeholders before provider routing.
CAN WE KEEP CLOSED MODELS OUT OF SPECIFIC WORKLOADS?
Yes. Enterprise policy can route sensitive work to a single-tenant open-weight deployment and reserve closed models for tasks where PII minimization is acceptable.
DO AUDIT LOGS STORE PROMPTS?
No. Audit logs record operational events such as user access, model invocation metadata, repository access, and policy changes. They are designed for governance without becoming another store of prompt content.